Image Origami

Image-based modeling systems use photographic images to ease the creation of and to enhance the realism of computer models. However, many of these systems require knowledge about camera parameters and 3D scene information, which may not be easily obtained. This paper proposes an interactive, image-based modeling system termed "Image Origami" that allows rapid construction of 3D environments from 2D images of architectural scenes without requiring a priori camera calibration. This method is based on reversing the 3D to 2D image projection model by allowing points marked in a 2D image to move along lines in 3-space defined by a 3D camera position and the point's projected 2D image plane location in 3space. The user specifies vertices, edges, and constraints, and can fold and bend images into 3D shapes based on these constraints. After constraints are entered, the system can automatically calculate the best-fit 3D approximation. The user is only required to enter information that is readily apparent, such as edge locations in an image and angles between edges and faces. To produce a more realistic texturemapped model, this system applies an occlusion removal technique.